| Most Recent Changes: | |
| Patch UVR_Patch_10_6_23_4_27: | |
| ~ MPS for MacOS is now compatible with all MDX-Net & VR Arch models! | |
| ~ Fixed memory issue with MDX23C models. | |
| ~ Added the ability to choose GPU device to process tracks. | |
| ~ Fixed a few graphical bugs. | |
| Other Changes: | |
| ~ Added "Vocal Split Mode," a chain ensemble that utilizes karaoke & BVE | |
| (backing vocal extractor) models to split vocals into lead vocal and backing vocal | |
| stems. | |
| ~ Updated "Audio Tools" to include "Matchering". | |
| ~ The "Align Tool" has been improved to line up inputs even when there are differences | |
| in timing, similar to what Utagoe does. | |
| ~ Integrated a right-click menu for entries in the batch process window specifically | |
| for Matchering & Align tool. | |
| ~ Introduced a right-click option for Matchering & Align tool on the main window to | |
| directly access input folders. | |
| ~ Addressed the anomaly where models would self-select in dropdowns. | |
| ~ This seemed related to having a multitude of models. | |
| ~ Revamped the dropdown menus entirely. | |
| ~ Fixed splash screen closing issue. | |
| ~ Demucs Pre-process Model settings properly reset to default | |
| ~ The MP3 encoder now uses LAME. | |
| ~ Resolved the problem of using the "MDX23C_D1581" model as a pre-process model for | |
| Demucs conversions were leading to a "Key Error: 'Primary Stem'" error. | |
| ~ Rectified the "missing file" error that was popping up when trying to convert using | |
| the Demucs v2 Demucs model. | |
| ~ Solved the problem in Demucs using the htdemucs_6s multi-stem model: when certain | |
| stems were selected and pitch-shift conversion was used, a KeyError prevented | |
| the stem from being saved. The stem separation now works correctly, and files are | |
| saved to the intended destination. | |
| ~ Adjusted the "Reset All Settings To Default" function to ensure it also resets the | |
| text of the Sample Mode slider. | |
| ~ Made corrections so that if you activate Model Test Mode and then restart the app, | |
| the "Accept Any Input" option will not be inadvertently enabled. | |
| ~ Addressed the ensemble issue with MDX23C models. | |
| ~ Updated the "Select Stems" menu for 2 stem models under MDX23C. | |
| ~ Revamped the MDXC23 Overlap Menu. | |
| ~ Provided an option to decide on using the VR denoise model in MDX-NET. | |
| ~ Introduced a selection menu with choices: "None," "Standard," and "Denoise Model." | |
| ~ Added an option to automatically deverb vocals for Demucs & MDX models. (does not | |
| work in ensemble mode at this time) | |
| ~ Updated the Download Center to feature MDX23C models. | |
| ~ Fixed the glitch where ensemble final outputs weren't saved. | |
| ~ Enhanced align/matchering outputs to support MP3 and FLAC. | |
| ~ Refined tooltips for "Overlap" and "Segment Default." | |
| ~ Ensured the download lists refresh properly even when there's no internet. | |
| ~ Many more fixes. | |
| Resources & Credits: | |
| ---------------------------------------------------------------------------------------------- | |
| Name: | |
| ZFTurbo | |
| Contribution: | |
| ~ Created the weights for the new MDX23C models. | |
| ~ Trained the new MDX23C models. | |
| Resources: | |
| ~ These models are available to use online. See link details below: | |
| ~ https://mvsep.com | |
| ~ Separation type -> MDX23C (vocals, instrumental) | |
| ~ Vocal model type -> 8K FFT, Full Band (SDR vocals: 10.17, SDR instrum: 16.48) | |
| ---------------------------------------------------------------------------------------------- | |
| Name: | |
| Bas Curtiz | |
| Contribution: | |
| ~ Conducted thorough testing and produced detailed tutorials on SDR evaluation. | |
| ~ Implemented systematic SDR quality assessments. | |
| ~ Authored Google Document detailing the optimal models for each genre. | |
| ~ Authored Google Document capturing UVR frequency cutoffs and time-elapsed comparisons between GPU and CPU performances. | |
| ~ Authored a Google Document consolidating best practices, tips, and tricks. | |
| Resources: | |
| ~ SDR Per Genre Breakdown | |
| ~ UVR Model Frequency Cutoff Chart | |
| ~ Tips & Tricks | |
| ~ Link: (See "Discord Resources Link" below) | |
| ---------------------------------------------------------------------------------------------- | |
| Name: | |
| deton24 | |
| Contribution: | |
| ~ SDR quality checks. | |
| ~ Authored a Google Document that serves as an all-in-one guide to source separation. | |
| Resources: | |
| ~ All-in-One Source Separation Guide | |
| ~ Link: (See "Discord Resources Link" below) | |
| ---------------------------------------------------------------------------------------------- | |
| Name: | |
| dca100fb8 | |
| Contribution: | |
| ~ Testing | |
| ~ Bug reporting | |
| ~ Suggestions | |
| ---------------------------------------------------------------------------------------------- | |
| Discord Resources Link: https://discord.com/channels/708579735583588363/1153421553694953482 | |